Skip to main content
NCBI home page
American Journal of Audiology logoLink to American Journal of Audiology
. 2022 Oct 10;31(4):1167–1177. doi: 10.1044/2022_AJA-21-00270

Predicting the Outcomes of Internet-Based Cognitive Behavioral Therapy for Tinnitus: Applications of Artificial Neural Network and Support Vector Machine

Hansapani Rodrigo a,b,✉, Eldré W Beukes b,c, Gerhard Andersson d,e, Vinaya Manchaiah b,f,g,h,i
PMCID: PMC9907438  PMID: 36215687

Abstract

Purpose:

Internet-based cognitive behavioral therapy (ICBT) has been found to be effective for tinnitus management, although there is limited understanding about who will benefit the most from ICBT. Traditional statistical models have largely failed to identify the nonlinear associations and hence find strong predictors of success with ICBT. This study aimed at examining the use of an artificial neural network (ANN) and support vector machine (SVM) to identify variables associated with treatment success in ICBT for tinnitus.

Method:

The study involved a secondary analysis of data from 228 individuals who had completed ICBT in previous intervention studies. A 13-point reduction in Tinnitus Functional Index (TFI) was defined as a successful outcome. There were 33 predictor variables, including demographic, tinnitus, hearing-related and treatment-related variables, and clinical factors (anxiety, depression, insomnia, hyperacusis, hearing disability, cognitive function, and life satisfaction). Predictive models using ANN and SVM were developed and evaluated for classification accuracy. SHapley Additive exPlanations (SHAP) analysis was used to identify the relative predictor variable importance using the best predictive model for a successful treatment outcome.

Results:

The best predictive model was achieved with the ANN with an average area under the receiver operating characteristic value of 0.73 ± 0.03. The SHAP analysis revealed that having a higher education level and a greater baseline tinnitus severity were the most critical factors that influence treatment outcome positively.

Conclusions:

Predictive models such as ANN and SVM help predict ICBT treatment outcomes and identify predictors of outcome. However, further work is needed to examine predictors that were not considered in this study as well as to improve the predictive power of these models.

Supplemental Material:

https://doi.org/10.23641/asha.21266487

Tinnitus is a prevalent hearing-related condition that affects 10%–15% of the adult population (McCormack et al., 2016). It is a highly heterogeneous condition in the way it is perceived and how individuals react to it. While tinnitus does not have an effect on the daily life of most people afflicted, a significant proportion reports that tinnitus severely affects the performance of their essential day-to-day tasks (Beukes et al., 2021). The common problems and life effects reported by individuals with tinnitus may include difficulties in emotional function, sleep, hearing, reduced energy levels, problems in handling stress, and social problems related to work and family (Beukes, Manchaiah, Andersson, et al., 2018; Manchaiah et al., 2018). Tinnitus is also associated with various clinical conditions such as hearing loss, anxiety, and depression. While there is no cure for tinnitus, there are various management strategies with varied empirical support (Tunkel et al., 2014).

According to most reviews and treatment guidelines, cognitive behavioral therapy (CBT) based on psychological principles has the most robust research support for tinnitus management (Fuller et al., 2020; Landry et al., 2020). CBT is based on the basic principle that what we think, how we feel, and how we behave are all closely connected, and each of these factors well-being. Due to this interconnectedness, addressing either unhelpful thoughts, emotions, reactions, or behaviors can lead to improvements associated with numerous disorders. CBT principles have been tailored as effective interventions for many different disorders including anxiety (Axelsson et al., 2019), depression (López-López et al., 2019), and insomnia (van der Zweerde et al., 2019), all which are often associated with tinnitus.

However, CBT is not easily accessible due to the limited number of trained professionals who can offer CBT for tinnitus (Bhatt et al., 2016). To increase the accessibility and affordability of CBT for tinnitus, an Internet-based cognitive behavioral therapy (ICBT) was developed as a self-help format of CBT with minimal guidance from therapists. Numerous advantages of this approach have expanded its applicability, including its accessibility and flexibility and being convenient and informative (Beukes, Manchaiah, Davies, et al., 2018). It furthermore requires less resources, having less associated stigma, partly due to maintaining anonymity and privacy and providing a standardized-evidence based treatment (Barak et al., 2008). As it is easily transferable, translatable, and culturally adaptable, it has the potential to reach global communities (Warmerdam et al., 2010).

As its efficacy and effectiveness have been demonstrated of other conditions associated with tinnitus, for example, insomnia (Yu et al. (2021), it was later also developed for tinnitus (Andersson et al., 2002). Several controlled trials across the globe have shown positive outcomes of ICBT for tinnitus (for a review, see the work of Beukes et al., 2019), but there is limited understanding of who will benefit from ICBT.

To identify for who ICBT is more suitable, studies have examined predictors of ICBT outcomes. Lower engagement has been found to be one of the main barriers of ICBT, estimated in about a quarter of all ICBT patients. Predictors of noncompliance are related to greater severity and higher levels of associated anxiety.

Factors including higher engagement and an increase in knowledge have led to larger reductions in severity for ICBT for insomnia (Kraepelien et al. 2021). Only a few previous studies have examined predictors of ICBT outcomes for tinnitus. Kaldo-Sandström et al. (2004) examined predictors of ICBT outcomes in a clinical population in Sweden. They reported that intervention compliance, external referral to the treatment, and the number of previous tinnitus treatments were associated with positive outcomes. In contrast, the number of messages sent to their therapist concerning the treatment problem was associated with worse outcomes. Further identified trends were that patients referred from external routes and those undertaking previous treatments had better outcomes.

When evaluating the long-term (1-year) outcomes of ICBT in the United Kingdom from a self-selected research population (Beukes, Allen, et al., 2018) found that baseline tinnitus severity, engagement with the ICBT program (more modules read), and higher self-reported satisfaction with the intervention were critical predictors of ICBT success. Rodrigo et al. (2021a) found that baseline tinnitus severity and education levels were predictors of ICBT outcome when applying univariate and multivariable models to a self-recruited research population undertaking ICBT in the United Kingdom. In a follow-up study, we used various exploratory techniques (Rodrigo et al., 2021b), namely, classification and regression trees (CART; Breiman et al., 1984), C5.0 decision trees (Quinlan, 1993), Gradient Boosting (Friedman, 2001), AdaBoost algorithm (Gandhi, 2018), eXtreme gradient Boosting (Chen & Guestrin, 2016), and Random Forest (Breiman, 2001). These models help to examine the presence of any nonlinear associations with the response. The CART decision tree model was identified as the optimal decision tree model. Nevertheless, its predictive power was still considered low (accuracy of 74%, sensitivity of 74%, specificity of 64%, and area under the receiver operating characteristic curve [AUC] of 0.69). This brought the necessity of exploring other nonlinear techniques to determine the most valuable predictors of outcome.

Artificial intelligence and machine learning (AI/ML) algorithms have become popular methods to predict outcomes in various rehabilitative practices. AI/MLs have been applied to audiovestibular data since the 1980s (Juhola et al., 2001). Several studies have reported that machine learning techniques such as artificial neural network (ANN) and support vector machine (SVM) have yielded more favorable results than other techniques for audiovestibular data (Haro et al., 2020; Niemann et al., 2020; Wang, 2017; Zhao et al., 2019).

ANN is an information processing archetype inspired by biological neural networks systems, such as the brain (McCulloch & Pitts, 1943). ANNs are typically organized in layers where each layer is made up of a number of interconnected “neurons” (also known as nodes), which contain an “activation function.” The purpose of the activation function is to introduce nonlinearity into the output of a neuron. This makes ANNs capable of capturing complex relationships intrinsic to the data. SVM (Cortes & Vapnik, 1995) is another well-known classifier that falls under supervised learning algorithms that uses the concept of “margin” to classify between two classes. SVM uses different kernels such as linear, polynomial, and radial basis to transform the inputs to higher dimensional feature space to assist in this task.

Few studies have applied ML to identify preintervention factors that may predict treatment outcomes for tinnitus. Niemann et al. (2020) applied AI techniques to determine treatment predictors for 1,416 patients completing a 7-day multimodal treatment for tinnitus. The authors used 205 predictor variables, including sociodemographic and clinical factors. As per their findings, gradient-boosted trees were identified to be the optimal classifier at predicting tinnitus-related distress postintervention. The variables, perceived tinnitus-related impairment, depressive symptoms, sleep problems, physical health-related impairments in quality of life, time spent to complete questionnaires, and educational level, exhibited a high contribution toward the model prediction. However, no study has used AI/ML to predict outcomes for ICBT tinnitus treatment other than our recent study (Rodrigo et al., 2021b).

This study aimed to examine the applications of AI/ML, specifically using ANN and SVM to identify variables associated with positive treatment outcomes in ICBT for tinnitus. First of all, we intend to identify the most appropriate AI/ML technique that can be used to predict treatment distress. This will be followed by identifying the most useful preintervention variables that can best predict postintervention outcomes.

Method

Study Design

The study was a secondary analysis of three different clinical trials that were used to investigate the impact of ICBT within the period from 2016 to 2018, namely, a single-group pretest, posttest design (n = 42; Beukes et al., 2017), an efficacy randomized control (RCT) design (n = 142; Clinical Trials.gov: NCT02370810; Beukes, Baguley, et al., 2018), and an effectiveness RCT design (n = 46; Clinical Trials.gov: NCT02665975; Beukes, Andersson, et al., 2018). Participants had completed the baseline and postintervention outcome measures, resulting in a pooled single data set (n = 228).

Ethical clearance was obtained from the Faculty of Science and Technology Research Ethics Panel of Anglia Ruskin University (ARU reference: FST/FREP/14/478 and FST/FREP/14/478) and the East of England–Cambridge South Research Ethics Committee (REC reference: 16/EE/0148) and Health Research Authority (IRAS project ID: 195565). All participants had provided informed consent before participating in the study.

Participant Characteristics

The inclusion criteria were age at least 18 years, ability to read and type in English, access to a computer and the Internet, and having bothersome tinnitus. The average age was 55.14 years (SD = 12.92), and 43% were women (see Supplemental Material S1). Participants had varied educational backgrounds, with 26% having completed high school education, 26% having an undergraduate degree, and 13% having a postgraduate degree. A quarter of the participants had sought tinnitus treatment previously, and the average tinnitus duration was 18 years (SD = 19 years)

Intervention

All participants had completed an 8-week ICBT intervention (Beukes et al., 2016, 2020). The intervention was presented in a self-help format with minimal therapist guidance and was administered using a secured ePlatform (Manchaiah et al., 2020). During this 8-week period, participants were represented with two to three learning modules that contained various elements of CBT specifically adapted for tinnitus, including applied relaxation, cognitive restructuring, and imagery. The digital assistive materials were presented using text, images, and videos. In addition, various exercises were presented to improve engagement.

Data Collection

All data were collected using online questionnaires. The study participants completed an extensive preintervention questionnaire, including demographics, tinnitus-related and treatment-related information, and other clinically relevant factors. Study participants also completed standardized patient-reported outcome measures before and after the intervention. The primary outcome was a change in tinnitus distress, as measured by the Tinnitus Functional Index (TFI; Meikle et al., 2012). The secondary outcome measures included the Insomnia Severity Index (Bastien et al., 2001) as a measure of insomnia, the Generalized Anxiety Disorder (Spitzer et al., 2006) as a measure of anxiety, the Patient Health Questionnaire (Spitzer et al., 1999) as a measure of depressive symptoms, the Hearing Handicap Inventory for Adults Screening version (Newman et al., 1991) as a measure of self-reported hearing disability, the Hyperacusis Questionnaire (Khalfa et al., 2002) to assess the presence of hyperacusis (reduced tolerance of everyday sounds), the Cognitive Failures Questionnaire (Broadbent et al., 1982) to evaluate cognitive functions, and the Satisfaction with Life Scales (Diener et al., 1985) to assess global life satisfaction.

Data Analysis

Variables

The dependent variable in the study was a change in tinnitus distress. A 13-point reduction in TFI scores was defined as a clinically significant change (Meikle et al., 2012), indicating successful treatment outcomes. In addition, there were 33 predictor variables, including seven demographic variables (age, gender, education level, employment type, noise exposure, the presence of psychological conditions, and tinnitus affecting the ability to work), 15 tinnitus and hearing-related variables (baseline tinnitus severity, tinnitus duration, how often tinnitus was heard, tinnitus location, nine different types of tinnitus, multiple tones heard, and the presence of hearing loss), four treatment-related variables (past treatment sought, tinnitus maskability [defined as sound enrichment being effective in making tinnitus less noticeable], hearing aid use, and medication use), and seven clinical factors (anxiety, depression, insomnia, hyperacusis, hearing disability, cognitive functions, and life satisfaction). Supplemental Materials S2–S5 provide details about these variables, including the specific questions and response options.

Preliminary analysis was performed to investigate differences in baseline characteristics between those who had shown clinically significant reduction in the TFI score versus those who had not, using either t test (for quantitative variables) or chi-square/Fishers' exact (for categorical variables) tests along with the odds ratio calculation. All analyses were two-tailed and performed with a 0.05 significance level. R statistical software (V.3.6.3) was used for all analyses, including predictive model development. The code and data are available upon a reasonable request.

Predictive Models

We focused on two types of machine learning models, namely, ANN (McCulloch & Pitts, 1943) and SVM (Cortes & Vapnik, 1995). The ANN model used in this study consisted of three layers: input, hidden, and output. As there were 33 predictor variables (taken as inputs to the ANN) and one outcome (taken as the output of the ANN, indicating the success or failure of the ICBT treatment), 33 input nodes and one output node were used in training the ANN model. A varying number of hidden nodes (from 1 to 5), and different weight decay values were examined during the ANN model training process. The optimal ANN model had five hidden nodes with a weight decay of 0.1. The sigmoid activation function was used at the output layer to obtain the predictions.

SVM employs kernel tricks and maximal margin concepts to perform better in nonlinear and high-dimensional tasks. Most of the time, even a powerful SVM model benefits from the proper feature selection and feature extraction/transformation techniques. Two SVM models were used in our study: one with linear kernel and one with Radial Basis Kernel (Chudzian, 2011).

Usually, traditional statistical methods are driven by certain distributional assumptions. If the model assumptions are violated, the conclusions made through these will not be valid. Moreover, when evaluating the effect of interactions within these models, those need to be added specifically and tested. Unlike these models, ML models such as ANN and SVM do not depend on the distribution of the data and can handle complex interactions within the data (Chong et al., 2021; Niu et al., 2019), even if they are not being specifically explored. More importantly, ANN and SVM are data-driven models, which have shown impressive predictive accuracies over other models (https://towardsdatascience.com/). However, ANN models are often criticized for their black-box nature as their predictions are nontransparent and not traceable by humans (Benitez et al., 1997). To overcome this, a model-agnostic post hoc framework, SHapley Additive exPlanations (SHAP), was applied to facilitate model interpretation and assess the feature importance when identifying the most influential factors leading to TFI score reduction after the treatment (Lundberg et al., 2019; Lundberg & Lee, 2017). Moreover, please note that, unlike traditional statistical methods, ANN models do not rely on model assumptions like multicollinearity (De Veaux & Ungar, 1994). Nevertheless, multicollinearity could affect the performance of SVM in a similar way to that of multicollinearity in logistic regression. Therefore, we have identified the variables that demonstrated severe multicollinearity using the generalized variation inflation factor (GVIF), with a multiple logistic regression model. Variables with VGIF > 10 indicate severe multicollinearity. Three variables, employment type (GVIF: 525.10), tinnitus location (GVIF: 15.73), and tinnitus affecting the ability to work (GVIF: 13.78), had shown severe multicollinearity. We ran two additional SVM models (one with linear and other with radial basis kernels) without these three variables, to investigate the impact of their removal on the model performance. Despite these changes, the discrimination power (as assessed by the mean AUC) of SVM models remain lower than the ANN model. Here, we summarize the predictive model development process step-by-step order to facilitate the readers' understanding.

Model Training

The data set with 228 participants was divided into two parts used for model training (80% of data, n = 183) and testing. During the model training process, a three-fold cross-validation process was applied. The training data set was split into three folds, and each fold was given a chance to act as its validation set to minimize model overfitting. Ten different replicate models were created with different random initializations for each AI method.

Model Performance Evaluation

The discriminative power of the trained models was assessed using the mean predictive accuracy, sensitivity (true positive rate), specificity (true negative rate), and AUC using the testing data set. These are given as M ± sd based on the 10 replicated models for each AI technique. The best predictive model was selected for further analysis based on the highest AUC value.

Shapley Additive exPlanations for Predictor Importance

To better understand the contribution of each predictor to the final model predictions, we used SHAP analysis (Lundberg & Lee, 2017). SHAP measures the impact of variables taking into account the interaction with other variables. SHAP estimates the importance of a predictor variable by comparing what a model predicts with and without that predictor variable. As the order in which a model sees variables can affect its predictions, every possible order was applied to ensure that the predictors were fairly compared.

For an overall comparison of predictor variables, a matrix of SHAP values was obtained for each participant. This matrix has one row per participant and one column per predictor, where we summed the absolute SHAP values for each predictor variable across the data. Predictors with large absolute SHAP values are considered globally significant. A visualization of the importance of these predictor variables in descending order is provided for the best predictive model.

Results

ICBT Outcome

Tinnitus distress was reduced from a mean of 58/100 points (SD = 19) to 34/100 points (SD = 23) after completing the ICBT intervention as measured by the TFI. This reduction was significant, t(227) = 16.37, p < .001. A clinically meaningful score change (13-point reduction in TFI) was achieved by 150/228 (66%) of the participants. We have performed a series of chi-square tests to evaluate this potential issue having higher TFI reduction mainly among the participants who had shown higher anxiety, higher depression, and higher insomnia (tinnitus-related distress factors at baseline). However, no association was found between TFI reduction and anxiety (OR = 1.09, 95% CI [0.60, 1.98], p = .774), between TFI reduction and depression (OR = 1.19, 95% CI [0.49, 2.88], p = .694), and between TFI reduction and insomnia (OR = 0.86, 95% CI [0.49, 1.50], p = .587). This indicates the fact that the effect of TFI reduction does not just associate with the participants with higher tinnitus-related distress measures. Supplemental Material S1 presents the t test or chi-square/Fisher's exact test results along with the corresponding odds ratios between the groups who had either shown clinically successful reduction or not.

Participants with a master's degree or above showed the highest odds ratio (OR = 4.50, 95% CI [1.59, 18.47], p = .003) of success with the ICBT treatment, compared with the participants who had an education only up to high school or less, indicating the importance of education. Baseline tinnitus severity was significantly different between the two groups (χ 1 = 53.27 [SD = 20.87], χ 2 = 60.35 (SD = 17.79), t(136.06) = −2.52, p = .012. Moreover, those for which tinnitus was not possible to mask was significantly different (OR = 3.31, 95% CI [1.02, 10.72], p = .042) between the groups and, in fact, indicates higher success compared with those who had full maskability. No other variables were identified as significantly different among the two groups. The findings align with the Rodrigo et al., (2021a, 2021b) studies, which had performed analysis on similar data using additional methods including linear and logistic regressions and decision tree models.

AI/ML Model Performance Evaluations

Results for the three predictive models examined (i.e., ANN, SVM with linear kernels, and SVM with radial basis kernels) are provided in Table 1. The ANN model led to the highest mean accuracy of 72.89 % (SD = 5.22) with 10 random initializations. The mean sensitivity and specificity values were 82.67 % (SD = 9.66) and 53.33 (SD = 9.94), respectively. The mean AUC value was 0.73 (SD = 0.03). The AUC is a measure of the discrimination capacity of the model, that is, the ability of the model to correctly identify those who might have a successful treatment outcome versus those who do not. A general rule of thumb is that an AUC between 0.90 and 1.00 indicates excellent discriminative power, an AUC between 0.8 and 0.9 indicates good discriminatory power, and an AUC between 0.70 and 0.80 indicates an adequate discriminative power. Hence, all three predictive models presented in this article show sufficient discriminative capacity. In contrast, ANN shows the highest average AUC out of all models considered here and in the work of Rodrigo et al. (2021b).

Table 1.

Predictive model evaluations.

Classification model Accuracy (%; SD) Sensitivity (true positive rate%; SD) Specificity (true negative rate%; SD) Area under the ROC curve (AUC%; SD)
ANN 72.89 ± 5.22 82.67 ± 9.66 53.33 ± 9.94 0.73 ± 0.03
SVM with linear kernel 69.12 ± 1.62 72.65 ± 4.40 61.99 ± 8.93 0.72 ± 0.01
SVM with linear kernel (without variables VIF > 10) 65.78 ± 3.95 84.00 ± 7.67 29.33 ± 25.77 0.68 ± 0.00
SVM with radial basis kernel 67.11 ± 4.54 74.33 ± 7.22 52.66 ± 2.10 0.70 ± 0.00
SVM with radial basis kernel (without variables VIF > 10) 66.66 ± 4.45 82.00 ± 5.71 35.99 ± 23.38 0.67 ± 0.01
Open in a new tab

Note. ROC = receiver operating characteristic; AUC = area under the receiver operating characteristic; ANN = artificial neural network; SVM = support vector machine; VIF = variation inflation factor.

Predictor Variable Importance With the Best ANN Predictive Model

Figure 1 shows the relative importance of each predictor variable toward the outcome variable based on the mean absolute SHAP values as identified by the optimal ANN model. These mean SHAP values represent the absolute change in log odds associated with a particular predictor variable. Hence, predictors with higher magnitude SHAP values indicate greater predictive power. As per the mean absolute SHAP values, education level, baseline tinnitus severity, presence of insomnia, employment type, and hearing aid usage were the top five predictive variables that are most influential toward the outcome. Following is a summary of individual predictor variable category effects for those top five predictor variables.

Figure 1.

Figure 1.

Open in a new tab

Feature importance based on mean absolute SHapley Additive exPlanation (SHAP) values from the optimal artificial neural network (ANN) model. TFI = Tinnitus Functional Index; HHIA = Hearing Handicap Inventory for Adults; CFQ = Cognitive Failures Questionnaire; SWLS = Satisfaction with Life Scales.

Demographic Variables

  • Higher education level (positive effect bachelor's degree and master's degree or above).

  • Age (positive effect for age > 57 years).

  • Employment type (positive effect professional, administrative, skilled tradesman, service occupation, medical, sales, home maker, and student).

Tinnitus-Related Variables

  • Greater baseline TFI score (positive effect with pre-TFI > 55.2/100).

  • Tinnitus location (positive effect for tinnitus in only one ear, unsure, and other categories).

  • Positive effect for participants who described their tinnitus as a buzzing sound.

  • Negative effect for participants who described their tinnitus as a clicking sound.

Treatment-Related Variables

  • Tinnitus maskability defined as sound enrichment being effective in making tinnitus less noticeable (positive effect for participants who never experienced tinnitus maskability at all or just partially).

  • Hearing aid use (positive effect for those who did not use a hearing aid or used an aid only in one ear).

Clinical Variables

  • Insomnia (positive effect with lower insomnia < 14/28)

Figure 2 shows the impact of each predictor variable category (feature effect) on the outcome. A positive SHAP value on a given feature, for the plot named “1” (where 1 indicates the group with successful treatment outcome) reflects a positive impact from that category on the treatment outcome, whereas a negative SHAP value indicates a negative effect on the outcome. Positive treatment effects were found for the groups who had higher education level (bachelor's degree and master's degree or above), and greater baseline tinnitus severity (pre-TFI > 55.2/100), lower insomnia (ISA < 14/28), employment types (professionals, administrative, skilled tradesmen, service occupational, medical, sales, homemaker, and students), hearing aid usage (using only one hearing aid or none), presence of buzzing type of tinnitus, maskability of tinnitus (defined as sound enrichment being effective in making tinnitus less noticeable), participants who never had tinnitus maskability at all or just partially, participants with age > 57 years, and tinnitus location (who had their tinnitus in only one ear, unsure, and other categories).

Figure 2.

Figure 2.

Open in a new tab

Effect of individual features on the outcomes according to the artificial neural network (ANN) model. Each graph represents a feature versus corresponding SHapley Additive exPlanation (SHAP) value. TFI = Tinnitus Functional Index; HHIA = Hearing Handicap Inventory for Adults; CFQ = Cognitive Failures Questionnaire; SWLS = Satisfaction with Life Scales.

Discussion

AI/ML approaches have not previously been applied to predict treatment outcomes. This study examined the applications of a few AI/ML-based models (ANN and SVM) in predicting the ICBT outcomes for tinnitus. Of the three models reviewed, the ANN model had the best predictive accuracy of postintervention improvement with a fair AUC value (0.73). The SVM models with linear kernel and also radial basis kernel had AUC values of 0.72 and 0.70, respectively. The ANN was also more accurate than other data mining models such as CART decision tree models (Rodrigo et al., 2021b). These findings align with several previous studies on audiovestibular data, identifying that ANN models are superior when examining audiovestibular data (Haro et al., 2020; Zhao et al., 2019). In contrast, a study examining the outcomes of a multimodal 7-day program for tinnitus found that the gradient-boosting model has the highest predictive accuracy (Niemann et al., 2020). The difference could be due to different predictor variables, differences in sample size, and different outcome measures.

Education level and baseline tinnitus severity have been consistently found to be the significant predictors throughout all our analyses (Rodrigo et al., 2021a, 2021b). Those with a master's education or higher and greater baseline tinnitus severity had shown high potential for ICBT success. This was expected as having good literacy skills is essential when understanding the intervention materials. The intervention materials used in these studies were written at an average of ninth-grade reading level, suggesting that they were not easily accessible for participants with only a high school education (Rodrigo et al., 2021a). The association of greater tinnitus severity for a successful ICBT treatment has been demonstrated earlier by Beukes, Allen, et al. (2018). More details on how other factors that have been identified in this study are related to ICBT treatment success can be found in the work of Rodrigo et al. (2021b). For other ICBT studies, greater severity and higher levels of associated anxiety have been found to be predictors. Higher intervention engagement and an increase in knowledge have been identified as additional predictors that reduce the severity of symptoms (Rozental et al., 2019), from which parallels can also be drawn. Similar results have also been obtained from ICBT for insomnia (Yeung et al., 2015) indicating greater insomnia, anxiety, and depression predicted noncompletion.

Although the ANN had the best predictive accuracy for this study, its discriminative capacity was only moderate (Swets, 1988). In general, a model giving an AUC value of 0.90 or above has high-discriminative power. Using a larger sample size and incorporating more relevant predictive variables might improve the predictive accuracy (Figueroa et al., 2012). Moreover, as the performance of models evaluated in this study and previous studies was similar (Niemann et al., 2020; Rodrigo et al., 2021b), it may be helpful to consider multiple models in future studies.

In addition to evaluating the predictive accuracy of data mining models, it is essential to examine which predictive variables contributed most to these models. In this study, education type and the baseline tinnitus severity were the most significant predictors of ICBT success. Most of these findings are consistent with previous studies that have used univariate, multivariable, and other data-mining techniques (Rodrigo et al., 2021a, 2021b). As ICBT is a self-help intervention, being able to read and follow written instructions is very important. For this reason, it is not surprising that those with a master's level of education or above had higher success. However, efforts should be made to improve the accessibility of ICBT, so that even those with limited education will receive good benefits. This can be done by revising the intervention for readability, as we have done in the recent U.S. version of the program (Beukes et al., 2020), as well as using a video-based intervention. Moreover, individuals with higher tinnitus distress (those with TFI scores of more than 50) are the ones who require more aggressive interventions such as the 8-week CBT program. For this reason, it is also not surprising that baseline tinnitus severity was the best predictor of ICBT success. However, efforts are needed to create different versions of the program to cater to individuals with varying levels of severity. In addition, the study was limited in terms of the number of predictive variables included. For this reason, future studies should aim to have other possible variables (e.g., health literacy, motivation, engagement, and adherence) that may influence the ICBT success.

In conclusion, this study has shown decision tree models such as ANN and SVM help predict ICBT treatment outcomes as well as in identifying the predictors of outcome. Further work is recommended using more variables and larger sample sizes to improve the predictive power of these models.

Data Availability Statement

The data can be made available upon reasonable request.

Supplementary Material

Supplemental Material S1. Characteristics of the study 740 participants and summary statistics for successful and unsuccessful treatment groups. Quantitative variables have been analyzed using a two-sample t test, while categorical data have been analyzed with chi-square or Fisher’s exact (denoted by an asterisk [*]). A threshold of .05 has been used.
Click here for additional data file. (527.7KB, pdf)
Supplemental Material S2. Demographic variables (7 variables).
Click here for additional data file. (424.1KB, pdf)
Supplemental Material S3. Tinnitus and hearing-related variables (15 variables).
Click here for additional data file. (503.3KB, pdf)
Supplemental Material S4. Treatment-related variables (4 variables).
Click here for additional data file. (415.7KB, pdf)
Supplemental Material S5. Clinical factors (7 variables).
Click here for additional data file. (509.8KB, pdf)

Acknowledgments

This work was partially funded by the National Institute on Deafness and Communication Disorders under Award Number R21DC017214 awarded to Vinaya Manchaiah.

Funding Statement

This work was partially funded by the National Institute on Deafness and Communication Disorders under Award Number R21DC017214 awarded to Vinaya Manchaiah.

References

  1. Andersson, G. , Strömgren, T. , Ström, L. , & Lyttkens, L. (2002). Randomized controlled trial of Internet-based cognitive behavior therapy for distress associated with tinnitus. Psychosomatic Medicine, 64(5), 810–816. https://doi.org/10.1097/01.psy.0000031577.42041.f8 [DOI] [PubMed] [Google Scholar]
  2. Axelsson, E. , & Hedman-Lagerlöf, E. (2019). Cognitive behavior therapy for health anxiety: Systematic review and meta-analysis of clinical efficacy and health economic outcomes. Expert Review of Pharmacoeconomics & Outcomes Research, 19(6), 663–676. https://doi.org/10.1080/14737167.2019.1703182 [DOI] [PubMed] [Google Scholar]
  3. Barak, A. , Hen, L. , Boniel-Nissim, M. , & Shapira, N. (2008). A comprehensive review and a meta-analysis of the effectiveness of Internet-based psychotherapeutic interventions. Journal of Technology in Human Service, 26(2-4), 109–160. https://doi.org/10.1080/15228830802094429 [Google Scholar]
  4. Bastien, C. H. , Vallières, A. , & Morin, C. M. (2001). Validation of the Insomnia Severity Index as an outcome measure for insomnia research. Sleep Medicine, 2(4), 297–307. https://doi.org/10.1016/s1389-9457(00)00065-4 [DOI] [PubMed] [Google Scholar]
  5. Benitez, J. M. , Castro, J. L. , & Requena, I. (1997). Are artificial neural networks black boxes? IEEE Transactions on Neural Networks, 8(5), 1156–1164. https://doi.org/10.1109/72.623216 [DOI] [PubMed] [Google Scholar]
  6. Beukes, E. W. , Allen, P. M. , Baguley, D. M. , Manchaiah, V. , & Andersson, G. (2018). Long-term efficacy of audiologist-guided Internet-based cognitive behavior therapy for tinnitus. American Journal of Audiology, 27(3S), 431–447. https://doi.org/10.1044/2018_AJA-IMIA3-18-0004 [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Beukes, E. W. , Allen, P. M. , Manchaiah, V. , Baguley, D. M. , & Andersson, G. (2017). Internet-based intervention for tinnitus: Outcome of a single-group open trial. Journal of the American Academy of Audiology, 28(4), 340–351. https://doi.org/10.3766/jaaa.16055 [DOI] [PubMed] [Google Scholar]
  8. Beukes, E. W. , Andersson, G. , Allen, P. M. , Manchaiah, V. , & Baguley, D. M. (2018). Effectiveness of guided Internet-based cognitive behavioral therapy vs face-to-face clinical care for treatment of tinnitus: A randomized clinical trial. JAMA Otolaryngology—Head & Neck Surgery, 144(12), 1126–1133. https://doi.org/10.1001/jamaoto.2018.2238 [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Beukes, E. W. , Baguley, D. M. , Allen, P. M. , Manchaiah, V. , & Andersson, G. (2018). Audiologist-guided internet-based cognitive behavior therapy for adults with tinnitus in the United Kingdom: A randomized controlled trial. Ear and Hearing, 39(3), 423–433. https://doi.org/10.1097/AUD.0000000000000505 [DOI] [PubMed] [Google Scholar]
  10. Beukes, E. W. , Fagelson, M. , Aronson, E. P. , Munoz, M. F. , Andersson, G. , & Manchaiah, V. (2020). Readability following cultural and linguistic adaptations of an Internet-based intervention for tinnitus for use in the United States. American Journal of Audiology, 29(2), 97–109. https://doi.org/10.1044/2019_AJA-19-00014 [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Beukes, E. W. , Manchaiah, V. , Allen, P. M. , Andersson, G. , & Baguley, D. M. (2021). Exploring tinnitus heterogeneity. Progress in Brain Research, 260, 79–99. https://doi.org/10.1016/bs.pbr.2020.05.022 [DOI] [PubMed] [Google Scholar]
  12. Beukes, E. W. , Manchaiah, V. , Allen, P. M. , Baguley, D. M. , & Andersson, G. (2019). Internet-based interventions for adults with hearing loss, tinnitus, and vestibular disorders: A systematic review and meta-analysis. Trends in Hearing, 23. https://doi.org/10.1177/2331216519851749 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Beukes, E. W. , Manchaiah, V. , Andersson, G. , Allen, P. M. , Terlizzi, P. M. , & Baguley, D. M. (2018). Situationally influenced tinnitus coping strategies: A mixed methods approach. Disability and Rehabilitation, 40(24), 2884–2894. https://doi.org/10.1080/09638288.2017.1362708 [DOI] [PubMed] [Google Scholar]
  14. Beukes, E. W. , Manchaiah, V. , Davies, A. S. , Allen, P. M. , Baguley, D. M. , & Andersson, G. (2018). Participants' experiences of an Internet-based cognitive behavioural therapy intervention for tinnitus. International Journal of Audiology, 57(12), 947–954. https://doi.org/10.1080/14992027.2018.1514538 [DOI] [PubMed] [Google Scholar]
  15. Beukes, E. W. , Vlaescu, G. , Manchaiah, V. , Baguley, D. M. , Allen, P. M. , Kaldo, V. , & Andersson, G. (2016). Development and technical functionality of an Internet-based intervention for tinnitus in the UK. Internet Interventions, 6, 6–15. https://doi.org/10.1016/j.invent.2016.08.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Bhatt, J. M. , Lin, H. W. , & Bhattacharyya, N. (2016). Prevalence, severity, exposures, and treatment patterns of tinnitus in the United States. JAMA Otolaryngology–Head & Neck Surgery, 142(10), 959–965. https://doi.org/10.1001/jamaoto.2016.1700 [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Breiman, L. (2001). Random forests. Machine Learning, 45, 5–32. https://doi.org/10.1023/A:1010933404324 [Google Scholar]
  18. Breiman, L. , Friedman, J. , Olshen, R. A. , & Stone, C. J. (1984). Classification and regression trees. Routledge. https://doi.org/10.1201/9781315139470 [Google Scholar]
  19. Broadbent, D. E. , Cooper, P. F. , FitzGerald, P. , & Parkes, K. R. (1982). The Cognitive Failures Questionnaire (CFQ) and its correlates. The British Journal of Clinical Psychology, 21(1), 1–16. https://doi.org/10.1111/j.2044-8260.1982.tb01421.x [DOI] [PubMed] [Google Scholar]
  20. Chen, T. , & Guestrin, C. (2016). XGBoost: A scalable tree boosting system. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. https://doi.org/10.1145/2939672.2939785 [Google Scholar]
  21. Chong, W. S. , Zaki, N. H. M. , Hossain, M. S. , Muslim, A. M. , & Pour, A. B. (2021). Introducing Theil-Sen estimator for sun glint correction of UAV data for coral mapping. Geocarto International. https://doi.org/10.1080/10106049.2021.1892206 [Google Scholar]
  22. Chudzian, P. (2011). Radial basis function kernel optimization for pattern classification. In Burduk R., Kurzyński M., Woźniak M., & Żołnierek A. (Eds.), Computer recognition systems 4. Advances in intelligent and soft computing (Vol. 95, pp. 99–108). Springer. https://doi.org/10.1007/978-3-642-20320-6_11 [Google Scholar]
  23. Cortes, C. , & Vapnik, V. (1995). Support-vector networks. Machine Learning, 20(3), 273–297. https://doi.org/10.1007/BF00994018 [Google Scholar]
  24. De Veaux, R. D. , & Ungar, L. H. (1994). Multicollinearity: A tale of two nonparametric regressions. In Cheeseman P. & Oldford R. W. (Eds.), Selecting models from data: AI and Statistics IV (pp. 393–402). Springer. https://doi.org/10.1007/978-1-4612-2660-4_40 [Google Scholar]
  25. Diener, E. , Emmons, R. A. , Larsen, R. J. , & Griffin, S. (1985). The Satisfaction with Life Scale. Journal of Personality Assessment, 49(1), 71–75. https://doi.org/10.1207/s15327752jpa4901_13 [DOI] [PubMed] [Google Scholar]
  26. Figueroa, R. L. , Zeng-Treitler, Q. , Kandula, S. , & Ngo, L. H. (2012). Predicting sample size required for classification performance. BMC Medical Informatics and Decision Making, 12. Article number 8. https://doi.org/10.1186/1472-6947-12-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Friedman, J. H. (2001). Greedy function approximation: A gradient boosting machine. Annals of Statistics, 29(5), 1189–1232. https://doi.org/10.1214/aos/1013203451 [Google Scholar]
  28. Fuller, T. , Cima, R. , Langguth, B. , Mazurek, B. , Vlaeyen, J. W. S. , & Hoare, D. J. (2020). Cognitive behavioural therapy for tinnitus. Cochrane Database of Systematic Reviews, 2020(1), CD012614. https://doi.org/10.1002/14651858.CD012614.pub2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Gandhi, R. (2018) Boosting algorithms: AdaBoost, Gradient Boosting and XGBoost. https://hackernoon.com/gradient-boosting-and-xgboost-90862daa6c77
  30. Haro, S. , Smalt, C. J. , Ciccarelli, G. A. , & Quatieri, T. F. (2020). Deep neural network model of hearing-impaired speech-in-noise perception. Frontiers in Neuroscience, 14, 588448. https://doi.org/10.3389/fnins.2020.588448 [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Juhola, M. , Viikki, K. , Laurikkala, J. , Auramo, Y. , Kentala, E. , & Pyykkö, I. (2001). Application of artificial intelligence in audiology. Scandinavian Audiology, 30(1), 97–99. https://doi.org/10.1080/010503901300007209 [DOI] [PubMed] [Google Scholar]
  32. Kaldo-Sandström, V. , Larsen, H. C. , & Andersson, G. (2004). Internet-based cognitive—Behavioral self-help treatment of tinnitus: Clinical effectiveness and predictors of outcome. American Journal of Audiology, 13(2), 185–192. https://doi.org/10.1044/1059-0889(2004/023) [DOI] [PubMed] [Google Scholar]
  33. Khalfa, S. , Dubal, S. , Veuillet, E. , Perez-Diaz, F. , Jouvent, R. , & Collet, L. (2002). Psychometric normalization of a hyperacusis questionnaire. ORL: Journal for Otorhinolaryngology and Its Related Specialties, 64(6), 436–442. https://doi.org/10.1159/000067570 [DOI] [PubMed] [Google Scholar]
  34. Kraepelien, M. , Blom, K. , Jernelöv, S. , & Kaldo, V. (2021). Weekly self-ratings of treatment involvement and their relation to symptom reduction in Internet cognitive behavioral therapy for insomnia. Cognitive Therapy and Research, 45(2), 262–271. https://doi.org/10.1007/s10608-020-10151-y [Google Scholar]
  35. Landry, E. C. , Sandoval, X. , Simeone, C. N. , Tidball, G. , Lea, J. , & Westerberg, B. D. (2020). Systematic review and network meta-analysis of cognitive and/or behavioral therapies (CBT) for tinnitus. Otology & Neurotology, 41(2), 153–166. https://doi.org/10.1097/MAO.0000000000002472 [DOI] [PubMed] [Google Scholar]
  36. López-López, J. A. , Davies, S. R. , Caldwell, D. M. , Churchill, R. , Peters, T. J. , Tallon, D. , Dawson, S. , Wu, Q. , Li, J. , Taylor, A. , Lewis, G. , Kessler, D. S. , Wiles, N. , & Welton, N. J. (2019). The process and delivery of CBT for depression in adults: A systematic review and network meta-analysis. Psychological Medicine, 49(12), 1937–1947. https://doi.org/10.1017/S003329171900120X [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Lundberg, S. M. , Erion, G. , Chen, H. , DeGrave, A. , Prutkin, J. M. , Nair, B. , Katz, R. , Himmelfarb, J. , Bansal, N. , & Lee, S.-I. (2019). Explainable AI for trees: From local explanations to global understanding. arXiv preprint arXiv:190504610 [DOI] [PMC free article] [PubMed]
  38. Lundberg, S. M. , & Lee, S.-I. (2017). A unified approach to interpreting model predictions. In Guyon I., Luxburg U. V., Bengio S., Wallach H., Fergus R., Vishwanathan S., & Garnett R. (Eds.), Advances in neural information processing systems 30 (pp. 4765–4774). Curran Associates, Inc. [Google Scholar]
  39. Manchaiah, V. , Beukes, E. W. , Granberg, S. , Durisala, N. , Baguley, D. M. , Allen, P. M. , & Andersson, G. (2018). Problems and life effects experienced by tinnitus research study volunteers: An exploratory study using the ICF classification. Journal of the American Academy of Audiology, 29(10), 936–947. https://doi.org/10.3766/jaaa.17094 [DOI] [PubMed] [Google Scholar]
  40. Manchaiah, V. , Vlaescu, G. , Varadaraj, S. , Aronson, E. P. , Fagelson, M. A. , Munoz, M. F. , Andersson, G. , & Beukes, E. W. (2020). Features, functionality, and acceptability of Internet-based cognitive behavioral therapy for tinnitus in the United States. American Journal of Audiology, 29(3), 476–490. https://doi.org/10.1044/2020_AJA-20-00002 [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. McCormack, A. , Edmondson-Jones, M. , Somerset, S. , & Hall, D. (2016). A systematic review of the reporting of tinnitus prevalence and severity. Hearing Research, 337, 70–79. https://doi.org/10.1016/j.heares.2016.05.009 [DOI] [PubMed] [Google Scholar]
  42. McCulloch, W. S. , & Pitts, W. (1943). A logical calculus of the ideas immanent in nervous activity. The Bulletin of Mathematical Biophysics, 5(4), 115–133. https://doi.org/10.1007/BF02478259 [PubMed] [Google Scholar]
  43. Meikle, M. B. , Henry, J. A. , Griest, S. E. , Stewart, B. J. , Abrams, H. B. , McArdle, R. , Myers, P. J. , Newman, C. W. , Sandridge, S. , Turk, D. C. , Folmer, R. L. , Frederick, E. J. , House, J. W. , Jacobson, G. P. , Kinney, S. E. , Martin, W. H. , Nagler, S. M. , Reich, G. E. , Searchfield, G. , … Vernon, J. A. (2012). The tinnitus functional index. Ear and Hearing, 33(2), 153–176. https://doi.org/10.1097/AUD.0b013e31822f67c0 [DOI] [PubMed] [Google Scholar]
  44. Newman, C. W. , Weinstein, B. E. , Jacobson, G. P. , & Hug, G. A. (1991). Test–retest reliability of the hearing handicap inventory for adults. Ear and Hearing, 12(5), 355–357. https://doi.org/10.1097/00003446-199110000-00009 [DOI] [PubMed] [Google Scholar]
  45. Niemann, U. , Boecking, B. , Brueggemann, P. , Mebus, W. , Mazurek, B. , & Spiliopoulou, M. (2020). Tinnitus-related distress after multi-modal treatment can be characterized using a key subset of baseline variables. PLOS ONE, 15(1), e0228037. https://doi.org/10.1371/journal.pone.0228037 [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Niu, W.-J. , Feng, Z.-K. , Feng, B.-F. , Min, Y.-W. , Cheng, C. T. , & Zhou, J. Z. (2019). Comparison of multiple linear regression, artificial neural network, extreme learning machine, and support vector machine in deriving operation rule of hydropower reservoir. Water, 11(1), 88. https://doi.org/10.3390/w11010088 [Google Scholar]
  47. Quinlan, J. R. (1993). C4.5: Programs for machine learning. Morgan Kaufmann Publishers, Inc. [Google Scholar]
  48. Rodrigo, H. , Beukes, E. W. , Andersson, G. , & Manchaiah, V. (2021a). Internet-based cognitive–behavioural therapy for tinnitus: Secondary analysis to examine predictors of outcomes. BMJ Open, 11(8), e049384. https://doi.org/10.1136/bmjopen-2021-049384 [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Rodrigo, H. , Beukes, E. W. , Andersson, G. , & Manchaiah, V. (2021b). Exploratory data mining techniques (Decision Tree Models) for examining the impact of Internet-based cognitive behavioral therapy for tinnitus: Machine learning approach. Journal of Medical Internet Research, 23(11), e28999. https://doi.org/10.2196/28999 [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Rozental, A. , Andersson, G. , & Carlbring, P. (2019). In the absence of effects: An individual patient data meta-analysis of non-response and its predictors in Internet-based cognitive behavior therapy. Frontiers in Psychology, 10, 589. https://doi.org/10.3389/fpsyg.2019.00589 [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Spitzer, R. L. , Kroenke, K. , & Williams, J. B. (1999). Validation and utility of a self-report version of PRIME-MD: The PHQ primary care study. Primary Care Evaluation of Mental Disorders. Patient Health Questionnaire. JAMA, 282(18), 1737–1744. https://doi.org/10.1001/jama.282.18.1737 [DOI] [PubMed] [Google Scholar]
  52. Spitzer, R. L. , Kroenke, K. , Williams, J. B. , & Lowe, B. (2006). A brief measure for assessing generalized anxiety disorder: The GAD-7. Archives of Internal Medicine, 166(10), 1092–1097. https://doi.org/10.1001/archinte.166.10.1092 [DOI] [PubMed] [Google Scholar]
  53. Swets, J. A. (1988). Measuring the accuracy of diagnostic systems. Science, 240(4857), 1285–1293. https://doi.org/10.1126/science.3287615 [DOI] [PubMed] [Google Scholar]
  54. Tunkel, D. E. , Bauer, C. A. , Sun, G. H. , Rosenfeld, R. M. , Chandrasekhar, S. S. , Cunningham, E. R., Jr. , Archer, S. M. , Blakley, B. W. , Carter, J. M. , Granieri, E. C. , Henry, J. A. , Hollingsworth, D. , Khan, F. A. , Mitchell, S. , Monfared, A. , Newman, C. W. , Omole, F. S. , Phillips, C. D. , Robinson, S. K. , … Whamond, E. J. (2014). Clinical practice guideline: Tinnitus. Otolaryngology—Head & Neck Surgery, 151(Suppl. 2), S1–S40. https://doi.org/10.1177/0194599814545325 [DOI] [PubMed] [Google Scholar]
  55. van der Zweerde, T. , Bisdounis, L. , Kyle, S. D. , Lancee, J. , & van Straten, A. (2019). Cognitive behavioral therapy for insomnia: A meta-analysis of long-term effects in controlled studies. Sleep Medicine Reviews, 48, 101208. https://doi.org/10.1016/j.smrv.2019.08.002 [DOI] [PubMed] [Google Scholar]
  56. Wang, D. (2017). Deep learning reinvents the hearing aid: Finally, wearers of hearing aids can pick out a voice in a crowded room. IEEE Spectrum, 54(3), 32–37. https://doi.org/10.1109/MSPEC.2017.7864754 [DOI] [PMC free article] [PubMed] [Google Scholar]
  57. Warmerdam, L. , Smit, F. , van Straten, A. , Riper, H. , & Cuijpers, P. (2010). Cost-utility and cost-effectiveness of Internet-based treatment for adults with depressive symptoms: Randomized trial. Journal of Medical Internet Research, 12(5), e53. https://doi.org/10.2196/jmir.1436 [DOI] [PMC free article] [PubMed] [Google Scholar]
  58. Yeung, W. F. , Chung, K. F. , Ho, F. Y. , & Ho, L. M. (2015). Predictors of dropout from Internet-based self-help cognitive behavioral therapy for insomnia. Behaviour Research and Therapy, 73, 19–24. https://doi.org/10.1016/j.brat.2015.07.008 [DOI] [PubMed] [Google Scholar]
  59. Yu, H. , Zhang, Y. , Liu, Q. , & Yan, R. (2021). Efficacy of online and face-to-face cognitive behavioral therapy in the treatment of neurological insomnia: A systematic review and meta-analysis. Annals of Palliative Medicine, 10(10), 10684–10696. https://doi.org/10.21037/apm-21-2387 [DOI] [PubMed] [Google Scholar]
  60. Zhao, Y. , Li, J. , Zhang, M. , Lu, Y. , Xie, H. , Tian, Y. , & Qiu, W. (2019). Machine learning models for the hearing impairment prediction in workers exposed to complex industrial noise: A pilot study. Ear and Hearing, 40(3), 690–699. https://doi.org/10.1097/AUD.0000000000000649 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplemental Material S1. Characteristics of the study 740 participants and summary statistics for successful and unsuccessful treatment groups. Quantitative variables have been analyzed using a two-sample t test, while categorical data have been analyzed with chi-square or Fisher’s exact (denoted by an asterisk [*]). A threshold of .05 has been used.
Click here for additional data file. (527.7KB, pdf)
Supplemental Material S2. Demographic variables (7 variables).
Click here for additional data file. (424.1KB, pdf)
Supplemental Material S3. Tinnitus and hearing-related variables (15 variables).
Click here for additional data file. (503.3KB, pdf)
Supplemental Material S4. Treatment-related variables (4 variables).
Click here for additional data file. (415.7KB, pdf)
Supplemental Material S5. Clinical factors (7 variables).
Click here for additional data file. (509.8KB, pdf)

Data Availability Statement

The data can be made available upon reasonable request.

Articles from American Journal of Audiology are provided here courtesy of American Speech-Language-Hearing Association

RESOURCES